The present invention relates to a method and apparatus for a pattern image processing with respect to image data which have been imaged by a scanning electron microscope (SEM), and more specifically to a pattern image processing apparatus and image processing method which are used for detecting a specified hole pattern within pattern images in which successive hole patterns are repeated on a very large scale integration circuit (VLSI) element and for simultaneously obtaining a diameter of the specified hole pattern.
The above-mentioned scanning electron microscope (SEM) is used for inspecting a hole pattern such as a contact hole which is formed on the VLSI element. This is to perform an image processing with respect to microscope image data which are imaged by the SEM, thereby to compare and detect as to whether or not the formed hole pattern is a desired shape by using a pattern recognition method.
There is a general method by a template matching method as a pattern recognition method in which a similar function is estimated to search the local maximum value and the global maximum value. In particular, there is well known a method for detecting a position at which a correlative coefficient becomes the maximum value by means of a fast Fourier transform (FFT) algorithm when the similar coefficient is a linear shape. In this method, a reference image (template) is previously registered, and the reference image and an object image are performed by two-dimensional FFT processing, respectively. A convolution between FFT processing images is performed, and then the convoluted processing images are performed by two dimensional inverse FFT processing to obtain an inverse FFT processing image. A pattern recognition is performed by detecting a position having the maximum value from values of respective position coordinates, namely correlative coefficients of the inverse FFT processing image.
In the case where a pattern recognition is performed by using an FFT algorithm, what pattern should be detected from repetitive patterns in the VLSI element depends on the objective image at that time. The problems occur with the above-mentioned condition as follows:
(1) The objective image has a pattern repetitively arranged as the same shape as the reference image, and a detected pattern depends on a condition for taking into the objective image because the correlative coefficient (a local maximum value) with each pattern has an extremely near value; PA1 (2) The pattern shape of the reference image does not necessarily have the sufficient features, thereby resulting in the case where a correlation failure occurs; PA1 (3) In the cases where the objective image has the convolution component, where the objective image has changes of scaling caused by the manufacturing step, and where the SEM image has a distortion, the correlation failure occurs; PA1 (4) Especially in an elliptic shape pattern, it is very rare for a revolution angle (for example, an angle formed by a beam scanning direction and a long diameter direction of the oval) of the images to coincide with between the reference image and the objective image, thereby reducing an accuracy of the pattern recognition because the local maximum value decreases; and
(5) It is desired that a magnification of the image increases with a pattern finer because a size of the previously registered reference image has the optimum value by a shape and size of the pattern to be performed the pattern recognition However, since the changes of the magnification corresponding to the pattern are unsuitable for a manufacturing process level, there is a problem that it is impossible to use a general processor.
Even though there is proposed a normalized correlation method as a method for solving the above problem, there is a problem that the general processor can not be used because the method has a huge calculation amount and is unsuitable for a processing time of the manufacturing process level except when the processor specified for the calculation is introduced.